15:02 < zoq> This should be the last test case for the recurrent network architecture; at least as long as somebody comes up with a neat idea for another test case.

15:04 < naywhayare> I need to read through all these test cases... I might learn something about neural networks :)

15:26 < zoq> http://www.mlpack.org/trac/wiki/NewStyleGuidelines

15:26 < zoq> JaraxussTong: I'll go an adjust the Train class so that it works with 2D and 1D data without explicitly specifying the dimension this week; Hopefully I can come up with a neat idea to easily build a network without using c++14 functionality.

15:26 < zoq> JaraxussTong: I tend to be picky about code, so maybe if your time permits you can look into the style guidelines: https://github.com/mlpack/mlpack/wiki/StyleGuidelines If not I'll go and do the modifications.

17:04 < zoq> Maybe someone has a neat idea: Right now the ann/trainer class provides an interface to handle 2D matrices as input; since the cnn network works on 3D matrices we need to modify some function, e.g. Train. Armadillo comes with some types to handle 3D matrices (cube/field). The problem is, we use col(...) to handle 2D data and slice(...) to handle 3D data.

17:05 < zoq> We can easily solve the problem by overloading the affected functions, but that results in some code duplication which I would ideally avoid. So maybe another solution is to use a function pointer which points to the right member function?

20:46 < stephentu> zoq: where'd you learn so much about neural nets